Currently there are two public o1 models
- o1-mini - excels at STEM, especially math and coding
- o1-preview - has strong reasoning capabilities and broad world knowledge.
In using ChatGPT’s two o1-models, I find that o1-preview excels at terminology and creating intuitive descriptions, outperforming previous models like 4o.
Over months of experience with various models, some users have developed specific workflows based on the strengths of each model. For instance, we often use a previous model to refine a prompt before leveraging the o1 model for its specialized capabilities.
When crafting prompts on unfamiliar topics, I usually ask the model to generate relevant terminology, which I then incorporate into my prompts. Although o1-preview is often superior at terminology, it’s sometimes excessive to use it solely for this purpose. This highlights the need for a simpler, faster, and more cost-effective o1 model tailored for terminology acquisition. Additionally, there’s value in an o1 model optimized for intuitive descriptions in cases where extensive reasoning isn’t necessary; this could be combined with the terminology model or developed as a separate one.